dynamics and energetics of solute permeation through the ... · dynamics and energetics of solute...
TRANSCRIPT
Dynamics and energetics of solute permeation through the
Plasmodium falciparum aquaglyceroporin
Camilo Aponte-Santamarıa,a Jochen S. Hubb and Bert L. de Groot*a
Received 16th March 2010, Accepted 15th June 2010
DOI: 10.1039/c004384m
The aquaglyceroporin from Plasmodium falciparum (PfAQP) is a potential drug target for the
treatment of malaria. It efficiently conducts water and other small solutes, and is proposed to
intervene in several crucial physiological processes during the parasitic life cycle. Despite the
wealth of experimental data available, a dynamical and energetic description at the single-
molecule level of the solute permeation through PfAQP has been lacking so far. Here we address
this question by using equilibrium and umbrella sampling molecular dynamics simulations.
We computed the water osmotic permeability coefficient, the pore geometry and the potential of
mean force for the permeation of water, glycerol and urea. Our simulations show that the PfAQP,
the human aquaporin 1 (hAQP1) and the Escherichia coli glycerol facilitator (GlpF) have nearly
identical water permeabilities. The Arg196 residue at the ar/R region was found to play a crucial
role regulating the permeation of water, glycerol and urea. The computed free energy barriers at
the ar/R selectivity filter corroborate that PfAQP conducts glycerol at higher rates than urea, and
suggest that PfAQP is a more efficient glycerol and urea channel than GlpF. Our results are
consistent with a solute permeation mechanism for PfAQP which is similar to the one established
for other members of the aquaglyceroporin family. In this mechanism, hydrophobic regions near
the NPA motifs are the main water rate limiting barriers, and the replacement of water–arg196
interactions and solute-matching in the hydrophobic pocket at the ar/R region are the main
determinants underlying selectivity for the permeation of solutes like glycerol and urea.
1. Introduction
The parasite Plasmodium falciparum is responsible for the
most lethal form of the malaria disease.1 It expresses one
aquaglyceroporin (PfAQP),2 that has been proposed to play
crucial physiological roles during the parasitic life cycle,2,3 and
is therefore a potential antimalarial drug target.4–6
PfAQP has been intensively studied experimentally over the
last decade. Functional studies2 established that PfAQP con-
ducts water and glycerol at high permeability rates, comparable
to the rates of the human aquaporin 1 (hAQP1) in the case
of water. These experiments, therefore, suggested that PfAQP
is implicated in tasks such as osmotic stress regulation, and
glycerol uptake for lipid synthesis and oxidative stress regula-
tion.2 In addition, PfAQP was found to be permeable to urea,2
ammonia,7 other polyols up to five carbons long,2 carbonyl
compounds3 and arsenite,2 suggesting that PfAQP may also be
involved in the release of toxic products. The biological role of
aquaglyceroporins for the Plasmodium parasites was further
investigated in experiments deleting the orthologue of PfAQP
in the rodent malaria parasite, Plasmodium berghei (PbAQP),
that showed a slower proliferation of the parasite in infected
mice.8 Furthermore, mutation studies demonstrated that a
glutamate residue located at the C loop (Glu125) near the
conserved Arg196 is critical for water permeation.9
The structure of PfAQP was recently determined by X-ray
crystallography at 2.05 A resolution.10 It revealed the 3D
architecture of the channel and provided molecular insights
into the efficient water and glycerol permeation. PfAQP has
the same fold as compared to other members of the family of
aquaglyceroporins.11 The pore geometry is remarkably similar
to the Escherichia coli glycerol facilitator (GlpF),10 the closest
homologous aquaglyceroporin (with a known structure) with
a sequence similarity of 50%.2 PfAQP arranges in a tetrameric
structure, where each monomeric unit constitutes a conduc-
tion pore. The conduction pore contains two constriction
regions: one located near the NPA motifs, that are replaced
in PfAQP by unusual NLA and NPS motifs, and a second
one at the ar/R region, where Arg196 is located facing the
hydrophobic Trp50 and Phe190 residues.10 In addition, the C
loop is anchored to the extracellular vestibule with Glu125
located in proximity to Arg196, highlighting on the impor-
tance of this region for the solute permeation,10 as previously
hypothesized from functional studies.9
Despite the wealth of data available, a dynamical and
energetic description at the single-molecule level of solute
permeation through PfAQP is lacking so far. Furthermore, a
systematic comparison of PfAQP with hAQP1 and GlpF (the
most studied aquaporin and aquaglyceroporin, respectively),
in terms of the solute permeation and their dynamical causes,
is incomplete. The goal of this paper is to address these two
questions employing molecular dynamics simulations, based
on the X-ray crystallographic structure of PfAQP. Initially,
the water permeation through PfAQP was quantified by
a Computational Biomolecular Dynamics Group, Max Planck Institutefor Biophysical Chemistry, Gottingen, Germany
bComputational and Systems Biology, Dept. of Cell and MolecularBiology, Uppsala University, Uppsala, Sweden
10246 | Phys. Chem. Chem. Phys., 2010, 12, 10246–10254 This journal is �c the Owner Societies 2010
PAPER www.rsc.org/pccp | Physical Chemistry Chemical Physics
computing the single molecule osmotic permeability coefficient.
Subsequently, the pore geometry and the energetics for water
transport through PfAQP were analyzed by computing the
radius and potential of mean force profiles from equilibrium
simulations. Finally the energetics of permeation of glycerol
and urea was studied by computing potentials of mean force
by using the technique of umbrella sampling simulations.
The paper is organized as follows: section 2.1 lists the
parameters and algorithms used for the equilibrium molecular
dynamics simulations; section 2.2 to 2.4 describes the methods
to derive the water permeability coefficients, the pore geometry
and potentials of mean force from equilibrium MD simula-
tions, and section 2.5 describes the umbrella sampling simula-
tions. The results are presented and discussed in sections 3 and
4. Finally, section 5 summarizes our conclusions.
2. Theory and methods
2.1 Equilibrium molecular dynamics simulations
Equilibrium molecular dynamics simulations were carried out
starting with the aquaporin tetramer in a fully solvated
Dipalmitoylphosphatidylcholine (DPPC) lipid bilayer (Fig. 1).
Three independent simulations of PfAQP aquaporin (labeled
from I to III) were carried out: a control simulation of the
wild-type form (I); a mutant where Glu125 (located at the C
loop) was mutated into serine (II), and a second mutant where
Arg196 (located at the pore face) was replaced by alanine (III).
The latter two simulations were performed to study the effect
of these two residues on the water permeation (Fig. 1(c)). In
addition, simulations of hAQP1 (IV) and GlpF (V) aqua-
porins were also carried out for comparison.
The simulation boxes contain the protein tetramer, 265
DPPC lipids (263 in simulation V) and around 21 500 SPC/E
water molecules.12 The PfAQP and GlpF structures were
taken from the Protein Data Bank (PDB ID codes 3C0210
and 1FX8,13 respectively). The starting structure of hAQP1
was modeled based on the X-ray structure of bovine AQP1
(PDB ID code 1J4N)14 by mutating differing residues by using
the WHAT IF modeling software.15 The tetramer was inserted
into the lipid bilayer by using the g_membed software.16
Crystallographic water molecules were kept in the structures
and ions were added to neutralize the simulation systems. The
amber99SB all-atom force field17 was used for the protein,
and lipid parameters were taken from Berger et al.18 The
simulations were carried out using the GROMACS simulation
software.19,20 Long-range electrostatic interactions were
calculated with the particle-mesh Ewald method.21,22 Short-
range repulsive and attractive interactions were described by a
Lennard-Jones potential, which was cut off at 1.0 nm. The
Settle algorithm23 was used to constrain bond lengths and
angles of water molecules and Lincs24 was used to constrain all
other bond lengths. The fastest angular degrees of freedom
involving hydrogen atoms were removed by using the virtual
interaction-sites algorithm,25 allowing a time step of 4 fs. The
temperature was kept constant by coupling the system to a
velocity rescaling thermostat26,27 at 300 K with a coupling
constant t = 0.5 ps. The pressure was kept constant by
coupling the system to a semiisotropic Parrinello-Rahman
barostat28 at 1 bar with a coupling constant of t = 5.0 ps.
All simulations were equilibrated for 4 ns before production.
During this equilibration time the coordinates of the protein
were harmonically restrained, with a harmonic force constant
of 1000 kJ mol�1/nm2. The simulation length of the produc-
tion runs was 100 ns discarding the first 10 ns as equilibration.
2.2 Water permeability calculations
The single-channel osmotic permeability, pf, was chosen to
quantify the water permeation through the simulated
Fig. 1 Molecular dynamics simulations of PfAQP. Side (a) and top
views (b) of the simulation box illustrating the tetramer (red, orange,
cyan and blue), fully embedded in a DPPC lipid bilayer (yellow
phosphorous atoms and green tails) and solvated by water (red,
white). (c) Conduction pore showing the helices HB and HE; the C
loop (cyan), and Glu125 and Arg196 (green). The blue lines indicate
the region considered to compute the water permeability coefficient pf.
This journal is �c the Owner Societies 2010 Phys. Chem. Chem. Phys., 2010, 12, 10246–10254 | 10247
aquaglyceroporins. A solute concentration difference, DCs,
between two media connected by a channel causes a net water
flux, j, through the channel, and the permeability coefficient,
pf, relates these two quantities:
j = �pfDCs. (1)
The pf was independently calculated for every monomer, based
on the collective diffusion model proposed by Zhu et al.29 A
collective variable n is defined as
dn ¼X
i
dzi=L; ð2Þ
where dzi are the displacements along the pore coordinate of
the water molecules inside the channel (labeled as i) during a
time dt, and L is the length of the channel. The variable n
obeys an Einstein relation,
hn2(t)i = 2Dnt, (3)
and the proportionality constant, 2Dn, is related with the
permeability coefficient:29
pf = uwDn, (4)
where uw C 30 A3 is the volume occupied by one water
molecule.
Water molecule displacements, dzi, were computed every
2 ps, within a cylindrical region (of length L = 2.05 nm and
radius r = 0.5 nm) centered at the pore axis and spanning
�0.7 nm down and 1.35 nm up from the NPA region
(Fig. 1(c)). hn2(t)i was obtained by averaging over 450 time
windows of 200 ps length each. Finally the pf was obtained
from the slope of the curve hn2(t)i versus time. An effective
pf value was obtained by averaging the values of the four
monomers and the error was estimated as the standard
error.
2.3 Pore dimensions
The geometry of the channels was monitored by computing
pore radius profiles with the HOLE software,30 averaging over
snapshots taken every 50 ps, yielding a standard error always
smaller than 0.4 A
2.4 Potential of mean force for water
The potential of mean force (PMF) for water was calculated to
identify rate limiting regions and relevant water–protein and
water–water interaction sites inside the pore. It was independently
calculated for every monomer by using the formula31
Gi(z) = �kBT lnhni(z)i, (5)
where kB is the Boltzmann constant, T is the simulation
temperature and hni(z)i is the average (over the whole trajectory)of the number of water molecules at the z position along the
ith pore. The number of water molecules is referred to a
cylinder (of radius 0.5 nm) that is aligned with the pore
coordinate (Fig. 1(c)). Therefore, the entropy at the bulk
regions is underestimated, and consequently, the bulk free
energy is also underestimated. To relate G(z) with the area of
one aquaporin monomer and account for such understima-
tions, a trapezoidal correction was applied in the entrance and
exit regions. In consequence, the final PMF refers to a density
of one channel per membrane area occupied by an aquaporin
monomer.32 The correction reads DGcyl = kBT ln(Amono/Ac),
where Amono denotes the membrane cross section area of the
aquaporin monomer and Ac the cross section area of the
cylinder (p � 0.25 nm2). The computed correction was
6.1 kJ/mol for PfAQP, 6.0 kJ/mol for hAQP1 and 6.2 kJ/mol
for GlpF. Prior to the analysis the protein monomers were
superimposed to a reference structure. The effective PMF,
Geff(z), was computed by combining the four monomer Gi(z) as
expð�GeffðzÞ=kBTÞ ¼1
4
X4
i¼1expð�GiðzÞ=kBTÞ
¼ 1
4
X4
i¼1hniðzÞi;
ð6Þ
and the error of Geff was estimated by propagating the standard
errors of hni(z)i, yielding a maximum uncertainty of approxi-
mately 1.5 kJ/mol.
2.5 Umbrella sampling simulations
The starting frames for the umbrella simulation were taken
from a 10 ns equilibrium simulations of PfAQP. The simula-
tion system contained the PfAQP tetramer, 264 POPE lipids,
20484 TIP4P water molecules,33 and four chloride ions.
Protein and glycerol interactions were described using the
OPLS all-atom force field,34,35 lipid parameters were taken
from from Berger et al.,18 and urea parameters from ref. 36
and 37. All simulation parameters were chosen as described in
the section 2.1, except that hydrogen atoms were not described
as virtual sites, requiring a time step of 2 fs. The PfAQP
channel was divided into 0.25 A (1 A for urea) wide equidistant
sections parallel to the membrane with the center of each
section representing an umbrella center. Subsequently, the
solute molecules (glycerol or urea) were placed into the
channel at the umbrella center. To enhance sampling, several
solute molecules were placed in each pore keeping a distance
of 20 A between the solute molecules. Water molecules that
overlapped with the solute were removed. The central carbon
atom of glycerol or urea was restrained in z direction by a
harmonic umbrella potential (k=1000 kJ mol�1/nm2). In addi-
tion, the solute molecules were restrained into cylinders that
were centered along the respective channel. Accordingly, a
flat-bottom quadratic potential in the xy-plane was applied on
the restrained atom (radius rc = 6 A, kc = 400 kJ mol�1/nm2).
After energy minimization, each system was simulated for 2 ns
(4 ns for urea PMFs).
After removing the first 500 ps (1 ns for urea) for equilibra-
tion from each umbrella simulation, 1440 umbrella histograms
(360 for urea) were extracted from the z-coordinate of the
restrained atom. Subsequently, the umbrella positions were
corrected with respect to the center of mass of the corres-
ponding monomer. This procedure avoids a possible unphysical
flattening of the PMF due to fluctuations of the PfAQP
monomers within the tetramer. Visual inspection of the umbrella
histograms showed that multiple histograms overlapped at
each z coordinate. The PMF was calculated using a cyclic
10248 | Phys. Chem. Chem. Phys., 2010, 12, 10246–10254 This journal is �c the Owner Societies 2010
implementation of the weighted histogram analysis method
(WHAM).38 The WHAM procedure incorporated the
integrated autocorrelation times (IACT) of the umbrella windows.
The IACTs were derived by fitting a double exponential to the
autocorrelation function of each window, allowing one to
analyically compute the IACTs. Because the IACT is subject
to large uncertainty in case of limited sampling (such as inside
the channel), we have subsequently smoothed the IACT along
the reaction coordinate using a moving average filter with a
width of 5 A.
Due to the cylindric flat-bottom restraint the umbrella
samplings yield a PMF which refers to a channel density of
one channel per cross section area of the cylinder. We also
corrected the PMF by a trapezoidal correction like the water
PMF above. In this case, Ac was chosen such that the entropy
of the solute in the flat-bottom quadratic potential equals the
entropy of a solute in a cylindrical well-potential of area Ac.
We found that this condition approximately holds if Ac is
computed via Ac = p(rc + 2sc)2 where sc = (kBT/kc)
1/2 is the
width of the Gaussian-shaped solute distribution at the edge of
the flat-bottom quadratic potential. Amono was estimated to
equal 10.7 nm2 and Ac was computed to 1.80 nm2, yielding a
correction of DGcyl = 4.4 kJ/mol. The statistical uncertainty of
the PMFs was estimated using bootstrap analysis as described
before,39 yielding an uncertainty of 2 kJ/mol (3 kJ/mol for
urea) at the main barrier and/or the main energy well.
3. Results
To quantify the water permeation through the different stu-
died aqua(glycero)porins, the permeability coefficient, pf, was
computed from equilibrium molecular dynamics simulations
(Fig. 2). The computed pf value of the wild type PfAQP is
3.3�0.2� 10�14 cm3/s. This value is nearly equal to the computed
value for hAQP1 and slighlty smaller than that of Glpf. In
addition, the mutation of Glu125 (at the C loop, Fig. 1) into
serine shows a reduction in the pf of 18% compared to the
wild type simulation. In contrast, the mutation of Arg196
(inside the narrow ar/R region of the pore, Fig. 1) into alanine
reveals a substantial increase (by a factor of two) in the water
permeability compared to the wild type simulation.
To identify the rate limiting regions in the pore that
determine the water permeation characteristics observed
above, the averaged pore radius and the PMF for water were
calculated (Fig. 3). In PfAQP, hAQP1 and GlpF, the pore
geometry remains essentially unchanged compared with the
hole profiles calculated from the X-ray structures,10 implying a
relatively rigid pore. In PfAQP (orange line) and GlpF
(cyan line) the pore profiles were found to be remarkably
similar, narrowing in the ar/R region down to 1.4 A radius,
and wider than the profile in hAQP1 (blue line). In addition,
the point mutation Glu125Ser (green line) does not modify the
pore geometry, whereas the Arg196Ala mutation (red line)
induces a widening of the pore to 1.7 A at the ar/R region.
The PMF for water permeation is also illustrated in Fig. 3.
In PfAQP (orange line), the PMF displays two major barriers
of 14.6 kJ/mol and 15.4 kJ/mol near the NPA region which is
not the most constricted region, and a third smaller barrier of
13.8 kJ/mol at the end of the ar/R region. In hAQP1 (blue line)
and GlpF (cyan line), the PMFs also have a main free energy
barrier at the NPA region, but they are smaller than the
barrier in PfAQP by 2.4 kJ/mol and 1.3 kJ/mol, respectively.
Furthermore, in PfAQP, the PMF displays eight local minima
inside the channel, and water molecules are predominantly
found at the positions of these minima. Consequently, the
effective water–water distance along the pore, computed as the
average distance between minima, is 2.6 � 0.3 A. At the same
region, the PMF for hAQP1 (GlpF) shows seven (eight)
minima separated by a water–water distance of 2.9 � 0.6 A
(2.8 � 0.6 A). In hAQP1, the water–water distance increment
is mainly observed between the NPA and the ar/R region.
Additionally, the Glu125Ser mutant (green line) does not
reveal significant changes in the PMF profile, whereas the
Arg196Ala mutant (red line) appears to have a complete
flattened profile at the ar/R region with a corresponding
decrease in the free energy barrier to a value less than
11 kJ/mol.
Finally, Fig. 4 depicts the PMFs for glycerol and urea,
derived from umbrella sampling simulations, for PfAQP
(orange line), and compared with previous molecular
dynamics studies for GlpF32 (cyan line). In PfAQP, the free
energy barrier at the ar/R region for glycerol is substantially
smaller than the existing for urea (approximately 21 kJ/mol
smaller). Moreover, for both solutes, glycerol and urea, the
free energy barrier at the ar/R region decreases in PfAQP
compared to that observed in GlpF. In addition, the PMF for
glycerol reveals new features in PfAQP that are not observed
in GlpF: first, there is a potential well at the intracellular
vestibule (z o �1.0 nm) instead of a free energy barrier.
Second, the potential wells at an intermediate region between
the intracellular vestibule and the NPA motifs are deeper in
PfAQP than in GlpF. Contrary to glycerol, the PMFs for urea
in PfAQP and GlpF were found to be similar between the
intracellular vestibule and the NPA motifs. Furthermore, for
both PfAQP and GlpF, glycerol crystallographic positions
(indicated with dots) agree favorably with the potential well
positions observed in the simulations.
Fig. 2 Single molecule water permeability coefficients derived from
equilibrium molecular dynamics simulations for the indicated proteins
and mutants.
This journal is �c the Owner Societies 2010 Phys. Chem. Chem. Phys., 2010, 12, 10246–10254 | 10249
4. Discussion
Here we present a computational study to characterize at the
molecular level the permeation of water, glycerol and urea
through the Plasmodium falciparum aquaglyceroporin
(PfAQP). Initially, the water permeation through PfAQP
was quantified by means of the single molecule osmotic
permeability coefficient, pf. Subsequently, the pore geometry
and the energetics for water, glycerol and urea transport
through PfAQP was analyzed by computing radius and poten-
tial of mean force profiles.
4.1 Quantifying the water transport activity
Let us first consider the absolute pf values predicted in our
simulations in light of permeation measurements and previous
computational studies. First, our computed values for the
reference (aquaporin) hAQP1 and (aquaglyceroporin) GlpF
(Fig. 2) are close to the experimental values, spanning between
4.6 � 10�14 cm3/s to 11.7 � 10�14 cm3/s for hAQP140–43 and
0.7 � 10�14 cm3/s for GlpF.44 Moreover, our predicted pf for
PfAQP is also close to these measurements. It should be noted
that there are no single-molecule pf measurements for PfAQP
reported so far to compare to directly. Second, our derived pfvalues are smaller than the values predicted in previous
computational studies on hAQP1 using the GROMOS force
field and the SPC water model31,45 by a factor of two.
Similarly, they are smaller than the values derived in studies
on hAQP1 and GlpF using the CHARMM force field and the
TIP3P water model46–48 at least by a factor of three. These
differences can be mainly attributed to the water diffusion
constant. For the SPCE water model (used in this study) the
diffusion constant is very close to the experimental value
(2.4 � 10�5 cm2/s). In contrast, for the SPC49 and the TIP3P33
water models (used in the mentioned previous computational
studies) the diffusion constant is overestimated by a factor of
1.5 and 2, respectively.
Let us now to consider the relative differences in the pf for
PfAQP compared to hAQP1 and GlpF. The computed pf for
PfAQP is close to the value predicted for hAQP1 (Fig. 2),
suggesting that PfAQP conducts water at a similar rates as
Fig. 3 Pore radius profiles (upper panels) and potential of mean force for water permeation profiles (lower panels) derived from equilibrium
molecular dynamics simulations. (a) Comparison between PfAQP (orange), hAQP1 (blue) and GlpF (cyan). (b) Effect of the point mutations
Glu125Ser (green) and Arg196Ala (red).
Fig. 4 PMFs for glycerol and urea permeation through PfAQP
(orange) compared with previous molecular dynamics studies for
GlpF32 (cyan). The dots correspond to the glycerol positions in the
PfAQP10 (orange dots) and Glpf13 (cyan dots) crystal structures.
10250 | Phys. Chem. Chem. Phys., 2010, 12, 10246–10254 This journal is �c the Owner Societies 2010
hAQP1. This result is in excellent agreement with experimental
data from oocyte swelling assays,2 and supports therefore that
PfAQP is a highly efficient water channel. Interestingly, the pffor GlpF was observed to be higher than the values for PfAQP
and hAQP1, suggesting that GlpF also conducts water at
appreciable rates (comparable to hAQP1). A water permeability
for GlpF has been observed in previous computational
studies,31,46–48 experimental assays with proteoliposomes50
and reconstituded planar bilayers,44 but with the experiments
showing a lower permeation rate than the computational
studies. The reason for this discrepancy remains unknown.
A possible explanation would be simulation inaccuracies.
However, given the excellent agreement between the simula-
tion and the experiment for AQP1, and the consistent results
for GlpF in several different and independent simulations, this
appears unlikely. Another possible explanation could lie in
the problematic estimation of the reconstitution efficiency,
required to derive the copy number to estimate the experi-
mantal single-channel pf.
4.2 The energetics of water permeation
Further calculations of the pore geometry and the PMF for
water (Fig. 3) allowed us to identify the rate limiting regions
that determine the water permeation properties presented
above. The averaged radius profiles demonstrate that PfAQP
and GlpF have practically identical pore geometries, wide
enough to allow the passage of water and other solutes such
as glycerol and urea. Accordingly, the narrower pore observed
for hAQP1 constitutes one of the main factors for the exclu-
sion of large solute molecules in this aquaporin. The average
pore geometry from the simulations is remarkably similar to
that from the initial crystal structures10,13,14 (all compared in
ref. 10), indicating a rigid channel.
The pore geometry itself is not sufficient to explain the water
transport rates but the full energetics must be taken into
account. The major rate limiting region (highest barriers in
the PMF) for PfAQP is observed at the NPA region, located
exactly at the same place as for hAQP1 and GlpF (Fig. 3). In
all three cases, two rings of hydrophobic residues sit there
(Fig. 5), and they constitute the main barriers for water
passage. This is in perfect agreement with previous simulations
of hAQP1 and GlpF,31,51 and confirms that PfAQP has similar
water regulation mechanisms as other members of the family
of aquaglyceroporins. Furthermore, in PfAQP, the Leu192
residue may force the permeating water molecules to interact
more strongly with the Asn residue than in GlpF (where this
Leu192 is replaced by Met202, a less hydrophobic residue,
Fig. 5), leading to the observed higher free energy barrier,
for PfAQP as compared to GlpF. Surprisingly, hAQP1 has
smaller free energy barriers than PfAQP and GlpF in this critical
region, despite the fact that the protein-water interactions are
much stronger due to a narrower pore and the presence of the
larger Phe24 and Phe56 residues in this region (Fig. 5).31
Finally, from the free energy barriers, it could be expected
that PfAQP, GlpF and hAQP1 conduct water at rates in the
following order: PfAQP o GlpF o hAQP1. Our pf calcula-
tions predict the lowest rate for PfAQP. However, hAQP1 did
not show the highest rate but GlpF did. This apparent contra-
diction between the pf and the PMF for hAQP1 and GlpF may
be due to the fact that a permeation event contributing to the
pf (the translocation of one effective water molecule from one
aqueous medium to the other) implies the collective motion of
all molecules inside the channel,29,52 and therefore multiple
free energy barriers and not only the highest should be taken
into account when relating permeation rates to permeation
energetics.
Fig. 5 Top view of the two hydrophobic rings (upper and lower panels) located near the NPA motifs in the indicated proteins.
This journal is �c the Owner Societies 2010 Phys. Chem. Chem. Phys., 2010, 12, 10246–10254 | 10251
4.3 The effect of point mutations
We investigated the role of Glu125 and Arg196 on the water
permeation of PfAQP. On one hand, the Glu125Ser mutant
shows neither a substantial reduction in the pf nor an
increment in the free energy barriers compared to the wild
type simulation (Fig. 2 and 3(b)), contrary to what previous
mutation experiments demonstrated.9 The reduction in the
water permeability due to the Glu125Ser mutation was attributed
to the destabilization of the C loop and further disruption of
the hydrogen bonds between Arg196 and Trp124.9,10 How-
ever, the conformational change of the C loop may occur on a
longer time scale than the simulated time (100 ns), impeding to
be detected in our equilibrium simulations. In fact, the C loop
was observed to be rigidly anchored to the extracellular
vestibule (with an average backbone rmsd of 0.97 A with a
standard deviation of 0.26 A) in the wild type simulation, and
just slightly more flexible (with an average backbone rmsd of
1.06 A and a standard deviation of 0.36 A) in the Glu125Ser
simulation.
On the other hand, the Arg196Ala mutant reveals a widening
of the pore at the Ar/R region and a corresponding reduction
in the free energy barrier to water permeation, leading to a
significant increase in the water permeability (Fig. 2 and 3(b)).
This result confirms that Arg196 is a crucial residue for the
water conduction through PfAQP.
4.4 Permeation of other solutes
We studied the energetics of glycerol and urea conduction
(Fig. 4). Let us first analyze the free energy barriers at the ar/R
region for the different simulated solutes. Our simulations
predict that PfAQP conducts urea at a lower rate than glycerol
which is in excellent agreement with oocyte swelling assays.2 In
PfAQP, like in other aquaglyceroporins, glycerol molecules
arrange at the ar/R in such a way that its hydroxyl groups can
interact with Arg196, replacing water–Arg196 hydrogen
bonds, and its apolar backbone orients towards the hydro-
phobic face (Trp50 and Phe190 residues).10,13 In constrast,
urea molecules cannot find this compensatory effect (dictated
by the amphiphilic nature of the pore), because increasing
the carbonyl–oxygen interactions with Arg196 would lead to
unfavourable orientations of the amine groups facing the
hydrophobic residues, and therefore would increase the free
energy barrier as indeed observed in our simulations.
Focussing at the differences between PfAQP and GlpF, it is
intriguing to note that PfAQP has a reduced free energy
barrier for both glycerol and urea compared to GlpF at the
ar/R region. This suggests that PfAQP is a more efficient
glycerol and urea channel than GlpF, despite the fact that
both have identical residues in this critical region. A possible
explanation for this would be that fewer water–arginine
hydrogen bonds need to be replaced by the permeating solute
molecule in the case of PfAQP compared to GlpF, and there-
fore the energetic cost to replace water with a solute at that
position is reduced. Consequently, in PfAQP, the arginine
residue instead would form more additional hydrogen bonds
with the surrounding residues (namely, Ser200 and Trp124)
than in GlpF, as proposed by Newby et al.10 Indeed, the
histograms of the number of hydrogen bonds between the
solvent and the arginine residue computed in the equilibrium
simulations (Fig. 6(a)) show on average fewer hydrogen bonds
for PfAQP than for GlpF, and accordingly, more hydrogen
bonds between the protein and the arginine residue for PfAQP
than for Glpf (Fig. 6(b)).
In addition, the glycerol PMF for PfAQP reveals two
features not observed in GlpF: a potential well at the intra-
cellular vestibule and a putative binding site inside the channel
near the NPA region This binding site is expected to affect the
water permeability, particularly, at high glycerol concentra-
tions. It is therefore highly interesting to see if this glycerol-
dependent water permeability can be observed experimentally.
5. Conclusions
Here we present the first molecular dynamics study to charac-
terize at the molecular level the permeation of water and
other solutes through the Plasmodium falciparum malaria
aquaglyceroporin (PfAQP). Our simulations confirm that
PfAQP is a highly efficient water channel, that is able to
conduct water at single-molecule permeability rates comparable
to the hAQP1 and GlpF rates. Furthermore, we identified the
Fig. 6 Water–Arg196 (a) and protein–Arg196 (b) hydrogen bond
distribution in PfAQP (orange), hAQP1 (blue) and GlpF (cyan).
10252 | Phys. Chem. Chem. Phys., 2010, 12, 10246–10254 This journal is �c the Owner Societies 2010
hydrophobic regions near the NPA motif as the main water
rate-limiting barriers, confirming therefore that PfAQP has a
similar water regulation mechanism as other members of the
family of aquaglyceroporins. Our simulations also support
that GlpF is a highly efficient water channel, with a water
permeability higher than that for hAQP1, as previously
observed in several different and independent simulation
studies.
We demonstrate that Arg196 located at the ar/R selectivity
filter plays a crucial role regulating the permeation of water
and other solutes such as glycerol or urea, as was previously
hypothesized from functional and crystallographic studies.
Our simulation results are consistent with a general permea-
tion mechanism for aquaglyceroporins in which water is rate-
limited by hydrophobic interactions in the NPA region. Other
solutes, such as glycerol or urea, must accommodate in the
ar/R selectivity filter, replacing water–arg196 hydrogen bonds
interactions and facing the hydrophobic Trp50 and Phe190
residues, a process that is favorable for glycerol but less
favorable for urea. In addition, in light of this mechanism,
our simulations suggest that PfAQP is able to conduct both
glycerol and urea at higher permeabilities than GlpF does,
because fewer water–Arg196 hydrogen bonds need to be
replaced by a permeating solute molecule in the case of PfAQP
compared to GlpF, a prediction to be further investigated and
validated in experimental studies.
The effect on the water permeability due to the mutation of
Glu125 (located at the C loop) to serine was found to be less
severe in our simulations than observed experimentally. This
could be attributed to the high stability of the C loop, rigidly
anchored to the extracellular vestibule, that would slow C loop
motions to time scales longer than the simulated time due to
the mentioned mutation.
This study is expected to guide further computational
and experimental studies in the search of putative blockers
of PfAQP, understanding of their mechanism of action, that
can hopefully be used to interrupt crucial physiological
processes of the malaria parasite such as the water regulation
and glycerol uptake.
Acknowledgements
We thank Erik Lindahl for providing us with the lipid
Berger-Amber parameters. We also thank Ulrike Gerischer
for carefully reading the manuscript. This work was supported
by grants from the European Commission, the Marie Curie
Research Training Network of Aqua(glycero)porins (MRTN-
CT-2006-035995), and by a Marie Curie Intra-European
Fellowship.
References
1 M. J. Gardner, N. Hall, E. Fung, O. White, M. Berriman,R. W. Hyman, J. M. Carlton, A. Pain, K. E. Nelson,S. Bowman, I. T. Paulsen, K. James, J. A. Eisen, K. Rutherford,S. L. Salzberg, A. Craig, S. Kyes, M.-S. Chan, V. Nene,S. J. Shallom, B. Suh, J. Peterson, S. Angiuoli, M. Pertea,J. Allen, J. Selengut, D. Haft, M. W. Mather, A. B. Vaidya, D.M. A. Martin, A. H. Fairlamb, M. J. Fraunholz, D. S. Roos,S. A. Ralph, G. I. McFadden, L. M. Cummings,G. M. Subramanian, C. Mungall, J. C. Venter, D. J. Carucci,
S. L. Hoffman, C. Newbold, R. W. Davis, C. M. Fraser andB. Barrell, Nature, 2002, 419, 498–511.
2 M. Hansen, J. A. F. J. Kun, J. E. Schultz and E. Beitz, J. Biol.Chem., 2002, 277, 4874–4882.
3 S. Pavlovic-Djuranovic, J. F. J. Kun, J. E. Schultz and E. Beitz,Biochim. Biophys. Acta, Biomembr., 2006, 1758, 1012–1017.
4 E. Beitz, ChemMedChem, 2006, 1, 587–592.5 E. Beitz, Biol. Cell, 2005, 97, 373–383.6 K. Kirk, Acta Trop., 2004, 89, 285–298.7 T. Zeuthen, B. Wu, S. Pavlovic-Djuranovic, L. M. Holm,N. L. Uzcategui, M. Duszenko, J. F. J. Kun, J. E. Schultz andE. Beitz, Mol. Microbiol., 2006, 61, 1598–1608.
8 D. Promeneur, Y. Liu, J. Maciel, P. Agre, L. S. King andN. Kumar, Proc. Natl. Acad. Sci. U. S. A., 2007, 104, 2211–2216.
9 E. Beitz, S. Pavlovic-Djuranovic, M. Yasui, P. Agre andJ. E. Schultz, Proc. Natl. Acad. Sci. U. S. A., 2004, 101, 1153–1158.
10 Z. E. R. Newby, J. O’Connell III, Y. Robles-Colmenares,S. Khademi, L. J. Miercke and R. M. Stroud, Nat. Struct. Mol.Biol., 2008, 15, 619–625.
11 T. Walz, Y. Fujiyoshi and A. Engel, Handbook of ExperimentalPharmacology: Aquaporins, 2009, pp. 31–56.
12 H. J. C. Berendsen, J. R. Grigera and T. P. Straatsma, J. Phys.Chem., 1987, 91, 6269–6271.
13 D. Fu, A. Libson, L. J. W. Miercke, C. Weitzman, P. Nollert,J. Krucinski and R. M. Stroud, Science, 2000, 290, 481–486.
14 H. Sui, B.-G. Han, J. K. Lee, P. Walian and B. K. Jap, Nature,2001, 414, 872–878.
15 G. Vriend, J. Mol. Graphics, 1990, 8, 52–56.16 M. G. Wolf, M. Hoefling, C. Aponte-Santamarıa, H. Grubmuller
and G. Groenhof, Journal of Computational Chemistry, 2010,http://dx.doi.org/10.1002/jcc.21507.
17 V. Hornak, R. Abel, A. Okur, B. Strockbine, A. Roitberg andC. Simmerling, Proteins: Struct., Funct., Bioinf., 2006, 65,712–725.
18 O. Berger, O. Edholm and F. Jahnig, Biophys. J., 1997, 72,2002–2013.
19 B. Hess, C. Kutzner, D. van der Spoel and E. Lindahl, J. Chem.Theory Comput., 2008, 4, 435–447.
20 D. V. D. Spoel, E. Lindahl, B. Hess, G. Groenhof, A. E. Mark andH. J. C. Berendsen, J. Comput. Chem., 2005, 26, 1701–1718.
21 U. Essmann, L. Perera, M. L. Berkowitz, T. Darden, H. Lee andL. G. Pedersen, J. Chem. Phys., 1995, 103, 8577–8593.
22 T. Darden, D. York and L. Pedersen, J. Chem. Phys., 1993, 98,10089–10092.
23 S. Miyamoto and P. A. Kollman, J. Comput. Chem., 1992, 13,952–962.
24 B. Hess, H. Bekker, H. J. C. Berendsen and J. G. E. M. Fraaije,J. Comput. Chem., 1997, 18, 1463–1472.
25 K. A. Feenstra, B. Hess and H. J. C. Berendsen, J. Comput. Chem.,1999, 20, 786–798.
26 G. Bussi, D. Donadio and M. Parrinello, J. Chem. Phys., 2007,126, 14101.
27 H. J. C. Berendsen, J. P. M. Postma, W. F. van Gunsteren,A. DiNola and J. R. Haak, J. Chem. Phys., 1984, 81, 3684–3690.
28 M. Parrinello and A. Rahman, J. Appl. Phys., 1981, 52, 7182–7190.29 F. Zhu, E. Tajkhorshid and K. Schulten, Phys. Rev. Lett., 2004, 93,
224501.30 O. S. Smart, J. M. Goodfellow and B. A. Wallace, Biophys. J.,
1993, 65, 2455–2460.31 B. L. de Groot and H. Grubmuller, Science, 2001, 294, 2353–2357.32 J. S. Hub and B. L. de Groot, Proc. Natl. Acad. Sci. U. S. A., 2008,
105, 1198–1203.33 W. L. Jorgensen, J. Chandrasekhar, J. D. Madura, R. W. Impey
and M. L. Klein, J. Chem. Phys., 1983, 79, 926–935.34 W. L. Jorgensen, D. S. Maxwell and J. Tirado-Rives, J. Am. Chem.
Soc., 1996, 118, 11225–11236.35 G. A. Kaminski, R. A. Friesner, J. Tirado-Rives and
W. L. Jorgensen, J. Phys. Chem. B, 2001, 105, 6474–6487.36 E. M. Duffy, D. L. Severance and W. L. Jorgensen, Israel Journal
of Chemistry, 1993, 33, 323–330.37 L. J. Smith, H. J. C. Berendsen and W. F. van Gunsteren, J. Phys.
Chem. B, 2004, 108, 1065–1071.38 S. Kumar, J. M. Rosenberg, D. Bouzida, R. H. Swendsen and
P. A. Kollman, J. Comput. Chem., 1992, 13, 1011–1021.39 J. S. Hub and B. L. de Groot, Biophys. J., 2006, 91, 842–848.
This journal is �c the Owner Societies 2010 Phys. Chem. Chem. Phys., 2010, 12, 10246–10254 | 10253
40 M. L. Zeidel, S. V. Ambudkar, B. L. Smith and P. Agre,Biochemistry, 1992, 31, 7436–7440.
41 M. L. Zeidel, S. Nielsen, B. L. Smith, S. V. Ambudkar,A. B. Maunsbach and P. Agre, Biochemistry, 1994, 33, 1606–1615.
42 T. Walz, B. L. Smith, M. L. Zeidel, A. Engel and P. Agre, Journalof Biological Chemistry, 1994, 269, 1583–1586.
43 B. Yang and A. S. Verkman, J. Biol. Chem., 1997, 272,16140–16146.
44 S. M. Saparov, S. P. Tsunoda and P. Pohl, Biol. Cell, 2005, 97,545–550.
45 B. L. de Groot and H. Grubmuller, Curr. Opin. Struct. Biol., 2005,15, 176–183.
46 F. Zhu, E. Tajkhorshid andK. Schulten,Biophys. J., 2002, 83, 154–160.
47 M. Jensen and O. G. Mouritsen, Biophys. J., 2006, 90, 2270–2284.48 M. Hashido, A. Kidera and M. Ikeguchi, Biophys. J., 2007, 93,
373–385.49 W. F. van Gusteren, J. Hermans, H. J. C. Berendsen and J. P.
M. Postma, Intermolecular forces, Reidel, Dordrecht, Holland,1981.
50 M. J. Borgnia and P. Agre, Proc. Natl. Acad. Sci. U. S. A., 2001,98, 2888–2893.
51 E. Tajkhorshid, P. Nollert, M. O. Jensen, L. J. W. Miercke,J. O’Connell, R. M. Stroud and K. Schulten, Science, 2002, 296,525–530.
52 F. Zhu, E. Tajkhorshid and K. Schulten, Biophys. J., 2004, 86,50–57.
10254 | Phys. Chem. Chem. Phys., 2010, 12, 10246–10254 This journal is �c the Owner Societies 2010